src/gromacs/gmxlib/nonbonded/nb_kernel_sparc64_hpc_ace_double/nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_sparc64_hpc_ace_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sparc64_hpc_ace_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_sparc64_hpc_ace_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sparc64_hpc_ace_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Particle-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_VF_sparc64_hpc_ace_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwjidx0A,vdwjidx0B;
  82     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  83     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  84     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
  85     real             *charge;
  86     int              nvdwtype;
  87     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  88     int              *vdwtype;
  89     real             *vdwparam;
  90     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
  91     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
  92     _fjsp_v2r8           c6grid_00;
  93     real                 *vdwgridparam;
  94     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
  95     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
  96     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
  97     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     real             *ewtab;
  99     _fjsp_v2r8       itab_tmp;
 100     _fjsp_v2r8       dummy_mask,cutoff_mask;
 101     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 102     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 103     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 104
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 117     charge           = mdatoms->chargeA;
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121     vdwgridparam     = fr->ljpme_c6grid;
 122     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 123     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 124     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 125
 126     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 129     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 130
 131     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 132     rcutoff_scalar   = fr->ic->rcoulomb;
 133     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 134     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 135
 136     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 137     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 138
 139     /* Avoid stupid compiler warnings */
 140     jnrA = jnrB = 0;
 141     j_coord_offsetA = 0;
 142     j_coord_offsetB = 0;
 143
 144     outeriter        = 0;
 145     inneriter        = 0;
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 163
 164         fix0             = _fjsp_setzero_v2r8();
 165         fiy0             = _fjsp_setzero_v2r8();
 166         fiz0             = _fjsp_setzero_v2r8();
 167
 168         /* Load parameters for i particles */
 169         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 170         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 171
 172         /* Reset potential sums */
 173         velecsum         = _fjsp_setzero_v2r8();
 174         vvdwsum          = _fjsp_setzero_v2r8();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185
 186             /* load j atom coordinates */
 187             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 192             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 193             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 197
 198             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 199
 200             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 201
 202             /* Load parameters for j particles */
 203             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 204             vdwjidx0A        = 2*vdwtype[jnrA+0];
 205             vdwjidx0B        = 2*vdwtype[jnrB+0];
 206
 207             /**************************
 208              * CALCULATE INTERACTIONS *
 209              **************************/
 210
 211             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 212             {
 213
 214             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 218             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 219                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 220
 221             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 222                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 223
 224             /* EWALD ELECTROSTATICS */
 225
 226             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 227             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 228             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 229             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 230             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 231
 232             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 233             ewtabD           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[1] );
 234             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 235             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 236             ewtabFn          = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[1] +2);
 237             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 238             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 239             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 240             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 241             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 242
 243             /* Analytical LJ-PME */
 244             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 245             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 246             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 247             exponent         = gmx_simd_exp_d(ewcljrsq);
 248             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 249             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 250             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 251             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 252             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 253             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 254                                _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
 255             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 256             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 257
 258             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 262             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 263             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 264             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 265
 266             fscal            = _fjsp_add_v2r8(felec,fvdw);
 267
 268             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 269
 270             /* Update vectorial force */
 271             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 272             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 273             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 274
 275             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 276
 277             }
 278
 279             /* Inner loop uses 79 flops */
 280         }
 281
 282         if(jidx<j_index_end)
 283         {
 284
 285             jnrA             = jjnr[jidx];
 286             j_coord_offsetA  = DIM*jnrA;
 287
 288             /* load j atom coordinates */
 289             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 290                                               &jx0,&jy0,&jz0);
 291
 292             /* Calculate displacement vector */
 293             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 294             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 295             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 296
 297             /* Calculate squared distance and things based on it */
 298             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 299
 300             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 301
 302             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 303
 304             /* Load parameters for j particles */
 305             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 306             vdwjidx0A        = 2*vdwtype[jnrA+0];
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 313             {
 314
 315             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 319             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 320                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 321
 322             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 323                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 329             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 330             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 331             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 332
 333             ewtabF           = _fjsp_load_v2r8( ewtab + 4*ewconv.i[0] );
 334             ewtabD           = _fjsp_setzero_v2r8();
 335             GMX_FJSP_TRANSPOSE2_V2R8(ewtabF,ewtabD);
 336             ewtabV           = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(), ewtab + 4*ewconv.i[0] +2);
 337             ewtabFn          = _fjsp_setzero_v2r8();
 338             GMX_FJSP_TRANSPOSE2_V2R8(ewtabV,ewtabFn);
 339             felec            = _fjsp_madd_v2r8(eweps,ewtabD,ewtabF);
 340             velec            = _fjsp_nmsub_v2r8(_fjsp_mul_v2r8(ewtabhalfspace,eweps) ,_fjsp_add_v2r8(ewtabF,felec), ewtabV);
 341             velec            = _fjsp_mul_v2r8(qq00,_fjsp_sub_v2r8(_fjsp_sub_v2r8(rinv00,sh_ewald),velec));
 342             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 343
 344             /* Analytical LJ-PME */
 345             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 346             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 347             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 348             exponent         = gmx_simd_exp_d(ewcljrsq);
 349             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 350             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 351             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 352             vvdw6            = _fjsp_mul_v2r8(_fjsp_madd_v2r8(c6grid_00,_fjsp_sub_v2r8(poly,one),c6_00),rinvsix);
 353             vvdw12           = _fjsp_mul_v2r8(c12_00,_fjsp_mul_v2r8(rinvsix,rinvsix));
 354             vvdw             = _fjsp_msub_v2r8(_fjsp_nmsub_v2r8(c12_00,_fjsp_mul_v2r8(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 355                                _fjsp_mul_v2r8(_fjsp_sub_v2r8(vvdw6,_fjsp_madd_v2r8(c6grid_00,sh_lj_ewald,_fjsp_mul_v2r8(c6_00,sh_vdw_invrcut6))),one_sixth));
 356             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 357             fvdw             = _fjsp_mul_v2r8(_fjsp_add_v2r8(vvdw12,_fjsp_msub_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6),vvdw6)),rinvsq00);
 358
 359             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _fjsp_and_v2r8(velec,cutoff_mask);
 363             velec            = _fjsp_unpacklo_v2r8(velec,_fjsp_setzero_v2r8());
 364             velecsum         = _fjsp_add_v2r8(velecsum,velec);
 365             vvdw             = _fjsp_and_v2r8(vvdw,cutoff_mask);
 366             vvdw             = _fjsp_unpacklo_v2r8(vvdw,_fjsp_setzero_v2r8());
 367             vvdwsum          = _fjsp_add_v2r8(vvdwsum,vvdw);
 368
 369             fscal            = _fjsp_add_v2r8(felec,fvdw);
 370
 371             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 372
 373             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 374
 375             /* Update vectorial force */
 376             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 377             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 378             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 379
 380             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 381
 382             }
 383
 384             /* Inner loop uses 79 flops */
 385         }
 386
 387         /* End of innermost loop */
 388
 389         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 390                                               f+i_coord_offset,fshift+i_shift_offset);
 391
 392         ggid                        = gid[iidx];
 393         /* Update potential energies */
 394         gmx_fjsp_update_1pot_v2r8(velecsum,kernel_data->energygrp_elec+ggid);
 395         gmx_fjsp_update_1pot_v2r8(vvdwsum,kernel_data->energygrp_vdw+ggid);
 396
 397         /* Increment number of inner iterations */
 398         inneriter                  += j_index_end - j_index_start;
 399
 400         /* Outer loop uses 9 flops */
 401     }
 402
 403     /* Increment number of outer iterations */
 404     outeriter        += nri;
 405
 406     /* Update outer/inner flops */
 407
 408     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*79);
 409 }
 410 /*
 411  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 412  * Electrostatics interaction: Ewald
 413  * VdW interaction:            LJEwald
 414  * Geometry:                   Particle-Particle
 415  * Calculate force/pot:        Force
 416  */
 417 void
 418 nb_kernel_ElecEwSh_VdwLJEwSh_GeomP1P1_F_sparc64_hpc_ace_double
 419                     (t_nblist                    * gmx_restrict       nlist,
 420                      rvec                        * gmx_restrict          xx,
 421                      rvec                        * gmx_restrict          ff,
 422                      struct t_forcerec           * gmx_restrict          fr,
 423                      t_mdatoms                   * gmx_restrict     mdatoms,
 424                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 425                      t_nrnb                      * gmx_restrict        nrnb)
 426 {
 427     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 428      * just 0 for non-waters.
 429      * Suffixes A,B refer to j loop unrolling done with double precision SIMD, e.g. for the two different
 430      * jnr indices corresponding to data put in the four positions in the SIMD register.
 431      */
 432     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 433     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 434     int              jnrA,jnrB;
 435     int              j_coord_offsetA,j_coord_offsetB;
 436     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 437     real             rcutoff_scalar;
 438     real             *shiftvec,*fshift,*x,*f;
 439     _fjsp_v2r8       tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 440     int              vdwioffset0;
 441     _fjsp_v2r8       ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 442     int              vdwjidx0A,vdwjidx0B;
 443     _fjsp_v2r8       jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 444     _fjsp_v2r8       dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 445     _fjsp_v2r8       velec,felec,velecsum,facel,crf,krf,krf2;
 446     real             *charge;
 447     int              nvdwtype;
 448     _fjsp_v2r8       rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 449     int              *vdwtype;
 450     real             *vdwparam;
 451     _fjsp_v2r8       one_sixth   = gmx_fjsp_set1_v2r8(1.0/6.0);
 452     _fjsp_v2r8       one_twelfth = gmx_fjsp_set1_v2r8(1.0/12.0);
 453     _fjsp_v2r8           c6grid_00;
 454     real                 *vdwgridparam;
 455     _fjsp_v2r8           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 456     _fjsp_v2r8           one_half = gmx_fjsp_set1_v2r8(0.5);
 457     _fjsp_v2r8           minus_one = gmx_fjsp_set1_v2r8(-1.0);
 458     _fjsp_v2r8       ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 459     real             *ewtab;
 460     _fjsp_v2r8       itab_tmp;
 461     _fjsp_v2r8       dummy_mask,cutoff_mask;
 462     _fjsp_v2r8       one     = gmx_fjsp_set1_v2r8(1.0);
 463     _fjsp_v2r8       two     = gmx_fjsp_set1_v2r8(2.0);
 464     union { _fjsp_v2r8 simd; long long int i[2]; } vfconv,gbconv,ewconv;
 465
 466     x                = xx[0];
 467     f                = ff[0];
 468
 469     nri              = nlist->nri;
 470     iinr             = nlist->iinr;
 471     jindex           = nlist->jindex;
 472     jjnr             = nlist->jjnr;
 473     shiftidx         = nlist->shift;
 474     gid              = nlist->gid;
 475     shiftvec         = fr->shift_vec[0];
 476     fshift           = fr->fshift[0];
 477     facel            = gmx_fjsp_set1_v2r8(fr->ic->epsfac);
 478     charge           = mdatoms->chargeA;
 479     nvdwtype         = fr->ntype;
 480     vdwparam         = fr->nbfp;
 481     vdwtype          = mdatoms->typeA;
 482     vdwgridparam     = fr->ljpme_c6grid;
 483     sh_lj_ewald      = gmx_fjsp_set1_v2r8(fr->ic->sh_lj_ewald);
 484     ewclj            = gmx_fjsp_set1_v2r8(fr->ic->ewaldcoeff_lj);
 485     ewclj2           = _fjsp_mul_v2r8(minus_one,_fjsp_mul_v2r8(ewclj,ewclj));
 486
 487     sh_ewald         = gmx_fjsp_set1_v2r8(fr->ic->sh_ewald);
 488     ewtab            = fr->ic->tabq_coul_F;
 489     ewtabscale       = gmx_fjsp_set1_v2r8(fr->ic->tabq_scale);
 490     ewtabhalfspace   = gmx_fjsp_set1_v2r8(0.5/fr->ic->tabq_scale);
 491
 492     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 493     rcutoff_scalar   = fr->ic->rcoulomb;
 494     rcutoff          = gmx_fjsp_set1_v2r8(rcutoff_scalar);
 495     rcutoff2         = _fjsp_mul_v2r8(rcutoff,rcutoff);
 496
 497     sh_vdw_invrcut6  = gmx_fjsp_set1_v2r8(fr->ic->sh_invrc6);
 498     rvdw             = gmx_fjsp_set1_v2r8(fr->ic->rvdw);
 499
 500     /* Avoid stupid compiler warnings */
 501     jnrA = jnrB = 0;
 502     j_coord_offsetA = 0;
 503     j_coord_offsetB = 0;
 504
 505     outeriter        = 0;
 506     inneriter        = 0;
 507
 508     /* Start outer loop over neighborlists */
 509     for(iidx=0; iidx<nri; iidx++)
 510     {
 511         /* Load shift vector for this list */
 512         i_shift_offset   = DIM*shiftidx[iidx];
 513
 514         /* Load limits for loop over neighbors */
 515         j_index_start    = jindex[iidx];
 516         j_index_end      = jindex[iidx+1];
 517
 518         /* Get outer coordinate index */
 519         inr              = iinr[iidx];
 520         i_coord_offset   = DIM*inr;
 521
 522         /* Load i particle coords and add shift vector */
 523         gmx_fjsp_load_shift_and_1rvec_broadcast_v2r8(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 524
 525         fix0             = _fjsp_setzero_v2r8();
 526         fiy0             = _fjsp_setzero_v2r8();
 527         fiz0             = _fjsp_setzero_v2r8();
 528
 529         /* Load parameters for i particles */
 530         iq0              = _fjsp_mul_v2r8(facel,gmx_fjsp_load1_v2r8(charge+inr+0));
 531         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 532
 533         /* Start inner kernel loop */
 534         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrA             = jjnr[jidx];
 539             jnrB             = jjnr[jidx+1];
 540             j_coord_offsetA  = DIM*jnrA;
 541             j_coord_offsetB  = DIM*jnrB;
 542
 543             /* load j atom coordinates */
 544             gmx_fjsp_load_1rvec_2ptr_swizzle_v2r8(x+j_coord_offsetA,x+j_coord_offsetB,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 549             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 550             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 556
 557             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = gmx_fjsp_load_2real_swizzle_v2r8(charge+jnrA+0,charge+jnrB+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562             vdwjidx0B        = 2*vdwtype[jnrB+0];
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 569             {
 570
 571             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 572
 573             /* Compute parameters for interactions between i and j atoms */
 574             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 575             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 576                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 577
 578             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 579                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 584             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 585             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 586             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 587             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 588
 589             gmx_fjsp_load_2pair_swizzle_v2r8(ewtab+ewconv.i[0],ewtab+ewconv.i[1],
 590                                          &ewtabF,&ewtabFn);
 591             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 592             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 593
 594             /* Analytical LJ-PME */
 595             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 596             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 597             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 598             exponent         = gmx_simd_exp_d(ewcljrsq);
 599             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 600             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 601             /* f6A = 6 * C6grid * (1 - poly) */
 602             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 603             /* f6B = C6grid * exponent * beta^6 */
 604             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 605             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 606             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 607
 608             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 609
 610             fscal            = _fjsp_add_v2r8(felec,fvdw);
 611
 612             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 613
 614             /* Update vectorial force */
 615             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 616             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 617             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 618
 619             gmx_fjsp_decrement_fma_1rvec_2ptr_swizzle_v2r8(f+j_coord_offsetA,f+j_coord_offsetB,fscal,dx00,dy00,dz00);
 620
 621             }
 622
 623             /* Inner loop uses 64 flops */
 624         }
 625
 626         if(jidx<j_index_end)
 627         {
 628
 629             jnrA             = jjnr[jidx];
 630             j_coord_offsetA  = DIM*jnrA;
 631
 632             /* load j atom coordinates */
 633             gmx_fjsp_load_1rvec_1ptr_swizzle_v2r8(x+j_coord_offsetA,
 634                                               &jx0,&jy0,&jz0);
 635
 636             /* Calculate displacement vector */
 637             dx00             = _fjsp_sub_v2r8(ix0,jx0);
 638             dy00             = _fjsp_sub_v2r8(iy0,jy0);
 639             dz00             = _fjsp_sub_v2r8(iz0,jz0);
 640
 641             /* Calculate squared distance and things based on it */
 642             rsq00            = gmx_fjsp_calc_rsq_v2r8(dx00,dy00,dz00);
 643
 644             rinv00           = gmx_fjsp_invsqrt_v2r8(rsq00);
 645
 646             rinvsq00         = _fjsp_mul_v2r8(rinv00,rinv00);
 647
 648             /* Load parameters for j particles */
 649             jq0              = _fjsp_loadl_v2r8(_fjsp_setzero_v2r8(),charge+jnrA+0);
 650             vdwjidx0A        = 2*vdwtype[jnrA+0];
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             if (gmx_fjsp_any_lt_v2r8(rsq00,rcutoff2))
 657             {
 658
 659             r00              = _fjsp_mul_v2r8(rsq00,rinv00);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq00             = _fjsp_mul_v2r8(iq0,jq0);
 663             gmx_fjsp_load_2pair_swizzle_v2r8(vdwparam+vdwioffset0+vdwjidx0A,
 664                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 665
 666             c6grid_00       = gmx_fjsp_load_2real_swizzle_v2r8(vdwgridparam+vdwioffset0+vdwjidx0A,
 667                                                                    vdwgridparam+vdwioffset0+vdwjidx0B);
 668
 669             /* EWALD ELECTROSTATICS */
 670
 671             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 672             ewrt             = _fjsp_mul_v2r8(r00,ewtabscale);
 673             itab_tmp         = _fjsp_dtox_v2r8(ewrt);
 674             eweps            = _fjsp_sub_v2r8(ewrt,_fjsp_xtod_v2r8(itab_tmp));
 675             _fjsp_store_v2r8(&ewconv.simd,itab_tmp);
 676
 677             gmx_fjsp_load_1pair_swizzle_v2r8(ewtab+ewconv.i[0],&ewtabF,&ewtabFn);
 678             felec            = _fjsp_madd_v2r8(eweps,ewtabFn,_fjsp_nmsub_v2r8(eweps,ewtabF,ewtabF));
 679             felec            = _fjsp_mul_v2r8(_fjsp_mul_v2r8(qq00,rinv00),_fjsp_sub_v2r8(rinvsq00,felec));
 680
 681             /* Analytical LJ-PME */
 682             rinvsix          = _fjsp_mul_v2r8(_fjsp_mul_v2r8(rinvsq00,rinvsq00),rinvsq00);
 683             ewcljrsq         = _fjsp_mul_v2r8(ewclj2,rsq00);
 684             ewclj6           = _fjsp_mul_v2r8(ewclj2,_fjsp_mul_v2r8(ewclj2,ewclj2));
 685             exponent         = gmx_simd_exp_d(ewcljrsq);
 686             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 687             poly             = _fjsp_mul_v2r8(exponent,_fjsp_madd_v2r8(_fjsp_mul_v2r8(ewcljrsq,ewcljrsq),one_half,_fjsp_sub_v2r8(one,ewcljrsq)));
 688             /* f6A = 6 * C6grid * (1 - poly) */
 689             f6A              = _fjsp_mul_v2r8(c6grid_00,_fjsp_sub_v2r8(one,poly));
 690             /* f6B = C6grid * exponent * beta^6 */
 691             f6B              = _fjsp_mul_v2r8(_fjsp_mul_v2r8(c6grid_00,one_sixth),_fjsp_mul_v2r8(exponent,ewclj6));
 692             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 693             fvdw              = _fjsp_mul_v2r8(_fjsp_madd_v2r8(_fjsp_msub_v2r8(c12_00,rinvsix,_fjsp_sub_v2r8(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 694
 695             cutoff_mask      = _fjsp_cmplt_v2r8(rsq00,rcutoff2);
 696
 697             fscal            = _fjsp_add_v2r8(felec,fvdw);
 698
 699             fscal            = _fjsp_and_v2r8(fscal,cutoff_mask);
 700
 701             fscal            = _fjsp_unpacklo_v2r8(fscal,_fjsp_setzero_v2r8());
 702
 703             /* Update vectorial force */
 704             fix0             = _fjsp_madd_v2r8(dx00,fscal,fix0);
 705             fiy0             = _fjsp_madd_v2r8(dy00,fscal,fiy0);
 706             fiz0             = _fjsp_madd_v2r8(dz00,fscal,fiz0);
 707
 708             gmx_fjsp_decrement_fma_1rvec_1ptr_swizzle_v2r8(f+j_coord_offsetA,fscal,dx00,dy00,dz00);
 709
 710             }
 711
 712             /* Inner loop uses 64 flops */
 713         }
 714
 715         /* End of innermost loop */
 716
 717         gmx_fjsp_update_iforce_1atom_swizzle_v2r8(fix0,fiy0,fiz0,
 718                                               f+i_coord_offset,fshift+i_shift_offset);
 719
 720         /* Increment number of inner iterations */
 721         inneriter                  += j_index_end - j_index_start;
 722
 723         /* Outer loop uses 7 flops */
 724     }
 725
 726     /* Increment number of outer iterations */
 727     outeriter        += nri;
 728
 729     /* Update outer/inner flops */
 730
 731     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*64);
 732 }